notebook
Writing
Notes on building, AI systems, experiments, and ideas I keep returning to.
Writing Index
Designing Tuku: from intent to execution
Why coding agents need an execution layer that can interpret intent, preserve continuity, and keep humans in control.
Why LLM Products Need Better Failure Loops
Most AI products optimize for first-response quality and ignore what happens after the model is wrong.
Building PebbleCode on AWS
Notes on why the platform ended up serverless, where the architecture bends, and what I would simplify next.
Making Model Context Protocol Understandable
What MCP Zero taught me about teaching a protocol through interfaces instead of documentation alone.
Building an Epistemic Audit Engine
A notebook entry on claim extraction, verification middleware, and why confidence scores need structure behind them.
What Hackathons Taught Me About Shipping AI Products
Fast constraints, rough prototypes, and the difference between a cool demo and a durable product idea.
Topics
Notes
Systems get interesting at the handoff points.
Most product quality is decided where one layer trusts another layer too early.
The right abstraction often feels obvious only after the interface exists.
A lot of product thinking is really interface research in disguise.
I trust metrics more when they change what the product does next.
Instrumentation is most useful when it feeds the next decision instead of becoming decoration.